File size: 16,255 Bytes
188f311 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 |
# Once for All: Train One Network and Specialize it for Efficient Deployment
# Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, Song Han
# International Conference on Learning Representations (ICLR), 2020.
import json
import torch.nn as nn
from proard.utils.layers import (
set_layer_from_config,
MBConvLayer,
ConvLayer,
IdentityLayer,
LinearLayer,
ResidualBlock,
)
from proard.utils import (
download_url,
make_divisible,
val2list,
MyNetwork,
MyGlobalAvgPool2d,
)
__all__ = ["proxyless_base_cifar","proxyless_base", "ProxylessNASNets", "MobileNetV2", "ProxylessNASNets_Cifar", "MobileNetV2_Cifar"]
def proxyless_base(
net_config=None,
n_classes=None,
bn_param=None,
dropout_rate=None,
local_path="~/.torch/proxylessnas/",
):
assert net_config is not None, "Please input a network config"
if "http" in net_config:
net_config_path = download_url(net_config, local_path)
else:
net_config_path = net_config
net_config_json = json.load(open(net_config_path, "r"))
if n_classes is not None:
net_config_json["classifier"]["out_features"] = n_classes
if dropout_rate is not None:
net_config_json["classifier"]["dropout_rate"] = dropout_rate
net = ProxylessNASNets.build_from_config(net_config_json)
if bn_param is not None:
net.set_bn_param(*bn_param)
return net
class ProxylessNASNets(MyNetwork):
def __init__(self, first_conv, blocks, feature_mix_layer, classifier):
super(ProxylessNASNets, self).__init__()
self.first_conv = first_conv
self.blocks = nn.ModuleList(blocks)
self.feature_mix_layer = feature_mix_layer
self.global_avg_pool = MyGlobalAvgPool2d(keep_dim=False)
self.classifier = classifier
def forward(self, x):
x = self.first_conv(x)
for block in self.blocks:
x = block(x)
if self.feature_mix_layer is not None:
x = self.feature_mix_layer(x)
x = self.global_avg_pool(x)
x = self.classifier(x)
return x
@property
def module_str(self):
_str = self.first_conv.module_str + "\n"
for block in self.blocks:
_str += block.module_str + "\n"
_str += self.feature_mix_layer.module_str + "\n"
_str += self.global_avg_pool.__repr__() + "\n"
_str += self.classifier.module_str
return _str
@property
def config(self):
return {
"name": ProxylessNASNets.__name__,
"bn": self.get_bn_param(),
"first_conv": self.first_conv.config,
"blocks": [block.config for block in self.blocks],
"feature_mix_layer": None
if self.feature_mix_layer is None
else self.feature_mix_layer.config,
"classifier": self.classifier.config,
}
@staticmethod
def build_from_config(config):
first_conv = set_layer_from_config(config["first_conv"])
feature_mix_layer = set_layer_from_config(config["feature_mix_layer"])
classifier = set_layer_from_config(config["classifier"])
blocks = []
for block_config in config["blocks"]:
blocks.append(ResidualBlock.build_from_config(block_config))
net = ProxylessNASNets(first_conv, blocks, feature_mix_layer, classifier)
if "bn" in config:
net.set_bn_param(**config["bn"])
else:
net.set_bn_param(momentum=0.1, eps=1e-3)
return net
def zero_last_gamma(self):
for m in self.modules():
if isinstance(m, ResidualBlock):
if isinstance(m.conv, MBConvLayer) and isinstance(
m.shortcut, IdentityLayer
):
m.conv.point_linear.bn.weight.data.zero_()
@property
def grouped_block_index(self):
info_list = []
block_index_list = []
for i, block in enumerate(self.blocks[1:], 1):
if block.shortcut is None and len(block_index_list) > 0:
info_list.append(block_index_list)
block_index_list = []
block_index_list.append(i)
if len(block_index_list) > 0:
info_list.append(block_index_list)
return info_list
def load_state_dict(self, state_dict, **kwargs):
current_state_dict = self.state_dict()
for key in state_dict:
if key not in current_state_dict:
assert ".mobile_inverted_conv." in key
new_key = key.replace(".mobile_inverted_conv.", ".conv.")
else:
new_key = key
current_state_dict[new_key] = state_dict[key]
super(ProxylessNASNets, self).load_state_dict(current_state_dict)
class MobileNetV2(ProxylessNASNets):
def __init__(
self,
n_classes=1000,
width_mult=1.0,
bn_param=(0.1, 1e-3),
dropout_rate=0.2,
ks=None,
expand_ratio=None,
depth_param=None,
stage_width_list=None,
):
ks = 3 if ks is None else ks
expand_ratio = 6 if expand_ratio is None else expand_ratio
input_channel = 32
last_channel = 1280
input_channel = make_divisible(
input_channel * width_mult, MyNetwork.CHANNEL_DIVISIBLE
)
last_channel = (
make_divisible(last_channel * width_mult, MyNetwork.CHANNEL_DIVISIBLE)
if width_mult > 1.0
else last_channel
)
inverted_residual_setting = [
# t, c, n, s
[1, 16, 1, 1],
[expand_ratio, 24, 2, 2],
[expand_ratio, 32, 3, 2],
[expand_ratio, 64, 4, 2],
[expand_ratio, 96, 3, 1],
[expand_ratio, 160, 3, 2],
[expand_ratio, 320, 1, 1],
]
if depth_param is not None:
assert isinstance(depth_param, int)
for i in range(1, len(inverted_residual_setting) - 1):
inverted_residual_setting[i][2] = depth_param
if stage_width_list is not None:
for i in range(len(inverted_residual_setting)):
inverted_residual_setting[i][1] = stage_width_list[i]
ks = val2list(ks, sum([n for _, _, n, _ in inverted_residual_setting]) - 1)
_pt = 0
# first conv layer
first_conv = ConvLayer(
3,
input_channel,
kernel_size=3,
stride=2,
use_bn=True,
act_func="relu6",
ops_order="weight_bn_act",
)
# inverted residual blocks
blocks = []
for t, c, n, s in inverted_residual_setting:
output_channel = make_divisible(c * width_mult, MyNetwork.CHANNEL_DIVISIBLE)
for i in range(n):
if i == 0:
stride = s
else:
stride = 1
if t == 1:
kernel_size = 3
else:
kernel_size = ks[_pt]
_pt += 1
mobile_inverted_conv = MBConvLayer(
in_channels=input_channel,
out_channels=output_channel,
kernel_size=kernel_size,
stride=stride,
expand_ratio=t,
)
if stride == 1:
if input_channel == output_channel:
shortcut = IdentityLayer(input_channel, input_channel)
else:
shortcut = ConvLayer(input_channel,output_channel,kernel_size=1,stride=1,bias=False,act_func=None)
else:
shortcut = None
blocks.append(ResidualBlock(mobile_inverted_conv, shortcut))
input_channel = output_channel
# 1x1_conv before global average pooling
feature_mix_layer = ConvLayer(
input_channel,
last_channel,
kernel_size=1,
use_bn=True,
act_func="relu6",
ops_order="weight_bn_act",
)
classifier = LinearLayer(last_channel, n_classes, dropout_rate=dropout_rate)
super(MobileNetV2, self).__init__(
first_conv, blocks, feature_mix_layer, classifier
)
# set bn param
self.set_bn_param(*bn_param)
def proxyless_base_cifar(
net_config=None,
n_classes=None,
bn_param=None,
dropout_rate=None,
local_path="~/.torch/proxylessnas/",
):
assert net_config is not None, "Please input a network config"
if "http" in net_config:
net_config_path = download_url(net_config, local_path)
else:
net_config_path = net_config
net_config_json = json.load(open(net_config_path, "r"))
if n_classes is not None:
net_config_json["classifier"]["out_features"] = n_classes
if dropout_rate is not None:
net_config_json["classifier"]["dropout_rate"] = dropout_rate
net = ProxylessNASNets_Cifar.build_from_config(net_config_json)
if bn_param is not None:
net.set_bn_param(*bn_param)
return net
class ProxylessNASNets_Cifar(MyNetwork):
def __init__(self, first_conv, blocks, feature_mix_layer, classifier):
super(ProxylessNASNets_Cifar, self).__init__()
self.first_conv = first_conv
self.blocks = nn.ModuleList(blocks)
self.feature_mix_layer = feature_mix_layer
self.global_avg_pool = MyGlobalAvgPool2d(keep_dim=False)
self.classifier = classifier
def forward(self, x):
x = self.first_conv(x)
for block in self.blocks:
x = block(x)
if self.feature_mix_layer is not None:
x = self.feature_mix_layer(x)
x = self.global_avg_pool(x)
x = self.classifier(x)
return x
@property
def module_str(self):
_str = self.first_conv.module_str + "\n"
for block in self.blocks:
_str += block.module_str + "\n"
_str += self.feature_mix_layer.module_str + "\n"
_str += self.global_avg_pool.__repr__() + "\n"
_str += self.classifier.module_str
return _str
@property
def config(self):
return {
"name": ProxylessNASNets_Cifar.__name__,
"bn": self.get_bn_param(),
"first_conv": self.first_conv.config,
"blocks": [block.config for block in self.blocks],
"feature_mix_layer": None
if self.feature_mix_layer is None
else self.feature_mix_layer.config,
"classifier": self.classifier.config,
}
@staticmethod
def build_from_config(config):
first_conv = set_layer_from_config(config["first_conv"])
feature_mix_layer = set_layer_from_config(config["feature_mix_layer"])
classifier = set_layer_from_config(config["classifier"])
blocks = []
for block_config in config["blocks"]:
blocks.append(ResidualBlock.build_from_config(block_config))
net = ProxylessNASNets_Cifar(first_conv, blocks, feature_mix_layer, classifier)
if "bn" in config:
net.set_bn_param(**config["bn"])
else:
net.set_bn_param(momentum=0.1, eps=1e-3)
return net
def zero_last_gamma(self):
for m in self.modules():
if isinstance(m, ResidualBlock):
if isinstance(m.conv, MBConvLayer) and isinstance(
m.shortcut, IdentityLayer
):
m.conv.point_linear.bn.weight.data.zero_()
@property
def grouped_block_index(self):
info_list = []
block_index_list = []
for i, block in enumerate(self.blocks[1:], 1):
if block.shortcut is None and len(block_index_list) > 0:
info_list.append(block_index_list)
block_index_list = []
block_index_list.append(i)
if len(block_index_list) > 0:
info_list.append(block_index_list)
return info_list
def load_state_dict(self, state_dict, **kwargs):
current_state_dict = self.state_dict()
for key in state_dict:
if key not in current_state_dict:
assert ".mobile_inverted_conv." in key
new_key = key.replace(".mobile_inverted_conv.", ".conv.")
else:
new_key = key
current_state_dict[new_key] = state_dict[key]
super(ProxylessNASNets_Cifar, self).load_state_dict(current_state_dict)
class MobileNetV2_Cifar(ProxylessNASNets_Cifar):
def __init__(
self,
n_classes=10,
width_mult=1.0,
bn_param=(0.1, 1e-3),
dropout_rate=0.2,
ks=None,
expand_ratio=None,
depth_param=None,
stage_width_list=None,
):
ks = 3 if ks is None else ks
expand_ratio = 6 if expand_ratio is None else expand_ratio
input_channel = 32
last_channel = 1280
input_channel = make_divisible(
input_channel * width_mult, MyNetwork.CHANNEL_DIVISIBLE
)
last_channel = (
make_divisible(last_channel * width_mult, MyNetwork.CHANNEL_DIVISIBLE)
if width_mult > 1.0
else last_channel
)
inverted_residual_setting = [
# t, c, n, s
[1, 16, 1, 1],
[expand_ratio, 24, 2, 1],
[expand_ratio, 32, 3, 2],
[expand_ratio, 64, 4, 2],
[expand_ratio, 96, 3, 1],
[expand_ratio, 160, 3, 2],
[expand_ratio, 320, 1, 1],
]
if depth_param is not None:
assert isinstance(depth_param, int)
for i in range(1, len(inverted_residual_setting) - 1):
inverted_residual_setting[i][2] = depth_param
if stage_width_list is not None:
for i in range(len(inverted_residual_setting)):
inverted_residual_setting[i][1] = stage_width_list[i]
ks = val2list(ks, sum([n for _, _, n, _ in inverted_residual_setting]) - 1)
_pt = 0
# first conv layer
first_conv = ConvLayer(
3,
input_channel,
kernel_size=3,
stride=1,
use_bn=True,
act_func="relu6",
ops_order="weight_bn_act",
)
# inverted residual blocks
blocks = []
for t, c, n, s in inverted_residual_setting:
output_channel = make_divisible(c * width_mult, MyNetwork.CHANNEL_DIVISIBLE)
for i in range(n):
if i == 0:
stride = s
else:
stride = 1
if t == 1:
kernel_size = 3
else:
kernel_size = ks[_pt]
_pt += 1
mobile_inverted_conv = MBConvLayer(
in_channels=input_channel,
out_channels=output_channel,
kernel_size=kernel_size,
stride=stride,
expand_ratio=t,
)
if stride == 1:
if input_channel == output_channel:
shortcut = IdentityLayer(input_channel, input_channel)
else:
shortcut = None #ConvLayer(input_channel,output_channel,kernel_size=1,stride=1,bias=False,act_func=None)
else:
shortcut = None
blocks.append(ResidualBlock(mobile_inverted_conv, shortcut))
input_channel = output_channel
# 1x1_conv before global average pooling
feature_mix_layer = ConvLayer(
input_channel,
last_channel,
kernel_size=1,
stride=1,
use_bn=True,
act_func="relu6",
ops_order="weight_bn_act",
)
classifier = LinearLayer(last_channel, n_classes, dropout_rate=dropout_rate)
super(MobileNetV2_Cifar, self).__init__(
first_conv, blocks, feature_mix_layer, classifier
)
# set bn param
self.set_bn_param(*bn_param)
|